Enhancing OpenMP Tasking Model: Performance and Portability

Abstract

OpenMP, as the de-facto standard programming model in symmetric multiprocessing for HPC, has seen its performance boosted continuously by the community, either through implementation enhancements or specification augmentations. Furthermore, the language has evolved from a prescriptive nature, as defined by the thread-centric model, to a descriptive behavior, as defined by the task-centric model. However, the overhead related to the orchestration of tasks is still relatively high. Applications exploiting very fine-grained parallelism and systems with a large number of cores available might fail on scaling. In this work, we propose to include the concept of Task Dependency Graph (TDG) in the specification by introducing a new clause, named taskgraph, attached to task or target directives. By design, the TDG allows alleviating the overhead associated with the OpenMP tasking model, and it also facilitates linking OpenMP with other programming models that support task parallelism. According to our experiments, a GCC implementation of the taskgraph is able to significantly reduce the execution time of fine-grained task applications and increase their scalability with regard to the number of threads.

Investigating Dependency Graph Discovery Impact on Task-based MPI+OpenMP Applications Performances

“…Data reuses increase on high TPL value. We measure a minimal time and work time at TPL=1,872 which is much bigger than the number of cores (24). The task-based version executes in 69𝑠 with MPC-OMP against 86𝑠 on the reference version using LLVM 16.…”

“…The standardisation of persistent task dependency graphs raises multiple issues. A taskgraph [24] proposal was recently made. However, it does not provide fine control on task attributes: everything is assumed persistent.…”

“…[35] does not report any timings on representative HPC application. In [24] the authors proposed an extension of the OpenMP tasking model to be able to capture the graph unroll by a code section in order to re-execute it on next iteration: all the closures are captured during first execution included firstprivate data passed by value. Because only simulated results are reported with no runtime implementation and experimental results, it is difficult to compare and evaluate the performance of the proposition.…”

Optimizing Iterative Data-flow Scientific Applications using Directed Cyclic Graphs

“…A task DAG caching proposal for OpenMP is the taskgraph clause for the target and task constructs [30]. Similar to CUDA Graphs, authors present a way to record and re-play task DAGs for OpenMP tasks.…”

“…4) Task Caching is the application adapted to simulate a task caching approach. We implemented the semantics of the taskgraph construct [30], where all data structures are cached between iterations, but without transformation or matching. In other words, it is a taskiter with a barrier between iterations.…”